BEGIN:VCALENDAR
VERSION:1.0
PRODID:Faculty of Science and Engineering - Research
BEGIN:VEVENT
SUMMARY:Chemistry Seminar: Prof Olexandr Ivanov and Prof Pavlo Prysyazhnyuk (Ivano-Frankivsk University, Ukraine)
DESCRIPTION;ENCODING=QUOTED-PRINTABLE: Talk title: "Towards Tungsten-Free Superhard Borides: an MLIP-Driven Crystal-Structure-Prediction Workflow with a DFT Calibration Gate, and Case Studies in Mn–Ti–B and Mn–V–B Systems"=0D=0A=
=0D=0A=
Abstract: =0D=0A=
=0D=0A=
Tungsten-based cemented carbides (WC–Co) have been the workhorse of cutting tools and wear-resistant coatings for nearly a century, but rising tungsten prices, EU and US critical-raw-material classifications, and tightening export controls now make the systematic search for tungsten-free alternatives a strategic=0D=0A=
priority for both industry and academic materials science. We present a computational workflow for the discovery of W-free superhard materials, based on machine-learned interatomic potentials (MLIPs) coupled with tensor-guided elastic-mode steering and a mandatory DFT validation gate. Methodology. The workflow combines random PyXtal generation with full 230-space-group enumeration, MLIP relaxation (MatterSim, GRACE-2L-OMAT), central finite-difference evaluation of the full 6 × 6 elastic stiffness tensor, and an iterative deformation strategy that steers candidate structures along the softest and hardest eigenmodes of the stiffness tensor to maximise Chen–Tian Vickers hardness while preserving low formation energy. The underlying mathematical apparatus — strain-tensor parametrisation, finite-difference scheme, eigenmode classification and adaptive step-size selection — will be presented in detail. Validation on known polymorphs. Across 10 chemically diverse benchmark systems (elemental C, Si, Ti, Zr; intermetallics=0D=0A=
Ni₃Al, TiAl; oxides MgO, SiO₂; borides TiB, WB), the workflow recovers 91 % of 222 known ICSD polymorphs through random generation alone, and an additional 2.7 % via mode-steering, with a ×2.48 enrichment in hard-phase hit frequency. A head-to-head benchmark against USPEX 25 on the B–N composition space recovers 6 / 11 stable phases (vs USPEX's 2 / 11) at 3.8×lower CPU-cost per phase, including the cubic Z-BN polymorph missed entirely by USPEX.=0D=0A=
=0D=0A=
Calibration and the phantom-hardness regime. Across 19 PBE-DFT validation jobs (ISIF = 2 relax + manual finite-difference elastic), MLIP-derived bulk moduli are accurate to ~6 %, but shear moduli and Chen–Tian hardness are systematically overestimated by 5–70× for low-symmetry, off-equilibrium phases — a "phantom-hardness" regime that two simple risk flags (residual pressure |P_res| > 30 kbar, RMS atomic displacement Δrms > 0.05 Å between MLIP- and DFT-relaxed structures) detect without any further DFT effort. We therefore propose a three-step validation gate (ISIF = 2 relax → finite-difference elastic → Born stability check) as a publication requirement for any MLIP-derived superhard claim.=0D=0A=
=0D=0A=
Case studies in W-free chemistries. Two manganese-based boride families illustrate the workflow on industrially relevant W-free chemistries. In Mn–Ti–B, six DFT-validated superhards have been identified, led by B₈MnTi₄ Fmm2 (Hv ≈ 46 GPa, ferromagnetic, dynamically stable, non-centrosymmetric — a candidate piezoelectric superhard), with the companion Mn–Cr–B system yielding the first altermagnetic boride identified through MLIP-driven CSP, B₈Cr₄Mn₄ Pnma (Hv = 24 GPa). In Mn–V–B, MLIP screening of ~10 000 candidates produces a Pareto front of low-Ef hard phases led by B₈MnV P2/m (Hv ≈ 44 GPa, Ef = −0.78 eV/atom) and the stability champion B₄MnV₃ (Hv ≈ 30 GPa, Ef = −1.02 eV/atom), now entering=0D=0A=
the DFT validation pipeline.=0D=0A=
=0D=0A=
Software implementation. The second half of the talk will detail the implementation — code architecture, parallelisation across MLIP backends, eigenmode-steering algorithm, persistent SQLite database design, and reproducibility features — enabling routine use of the workflow on commodity HPC partitions.=0D=0A=
=0D=0A=
 =0D=0A=
=0D=0A=
ABOUT THE SPEAKERS=0D=0A=
Pavlo Prysyazhnyuk — Professor, Doctor of Science, Department of Computerised Mechanical Engineering, Ivano-Frankivsk National Technical University of Oil and Gas (IFNTUNG), Ivano-Frankivsk, Ukraine.=0D=0A=
=0D=0A=
Computational materials scientist working on superhard and ultra-incompressible borides, MLIP-driven crystal structure prediction, and brazing / cladding alloy design.=0D=0A=
=0D=0A=
- University page: link here=0D=0A=
- Google Scholar:  link here=0D=0A=
- ResearchGate:    link here=0D=0A=
=0D=0A=
Olexandr Ivanov — PhD, Associate Professor and Head of the Department of Information Technologies, King Danylo University (UKD), Ivano-Frankivsk, Ukraine.=0D=0A=
=0D=0A=
Computer scientist working on modular software architecture, workflow automation for materials design, and parallel performance optimization.=0D=0A=
=0D=0A=
- University page: link here=0D=0A=
- Google Scholar:  link here=0D=0A=
- ResearchGate:    link here
LOCATION:G. O. Jones LT
DTSTART:20260608T140000
DTEND:20260608T153000
END:VEVENT
END:VCALENDAR